201101841 六、發明說明： 【發明所屬之技術領域】 本發明係關於頻率提高（Frame Rate Up-Conversion， FRC)技術’特別是關於一種自適應移動補償内插法 (Adaptive blending MCI)與其裝置。 【先前技術】 畫面頻率提高技術（frame rate up-conversion, FRC) o ^ 係用來增加視訊源（video source)的畫面頻率，一般業界有 各式各樣的應用’例如：應用於低位元的網路視訊傳輸以 節省頻寬（bandwidth)、應用於將畫面頻率25Hz的視訊源 轉換為更咼晝面頻率以減少畫面抖動（juddering)現象、及 應用於保持驅動型（hold-type)液晶顯示器（liquid crystal display ’ LCD)以避免晝面的模糊（blurring)現象並得到更 清晰的影像品質。 大部分的畫面頻率提高技術係先利用移動估計 〇 (motion estimation，ME)來計算出物體的移動向量，再進 行移動補償内插（motion compensation interpolation，MCI) 來内插出位於不同圖框（frame)的移動物體。大部分的移動 估計法都是以區塊的方式來進行，所以移動補償内插技術 的主要目的之一就是減少因區塊内插所造成的塊狀雜訊 (block artifacts)，先前的技術利用周邊區塊混成的方式去 減低塊狀雜訊，但是一般的方法並不區分週邊區塊是否正 破，有些方法以絕對誤差總和（sum of absolute 201101841 difference，SAD)的倒數值對週邊區塊加以選擇但是計 算式過於複雜，也不能正確反應週邊區塊的正確性。 【發明内容】 本發明一實施例&供了 一種移動補償内插裝置，包含有 一權重值單元與一混成單元。該權重值單元係接收至少一 圖框，該圖框包含複數個區塊，且於預設數目之該些區塊 〇 中規劃出一内插區塊，其中内插區塊包含至少—欲内插之 目標像素點，且接收内插區塊相鄰之複數個區塊對應之複 數個向量，測試該些向量，依據通過測試之至少一向量計 算出複數個權重值。而混成單元係接收複數個向量所對應 的複數個像素值，根據複數個權重值與複數個像素値產生 該欲内插之目標像素點之像素值。 本發明一實施例提供一種移動補償内插方法，包含有下 列步驟：首先，於一圖框規劃一内插區塊，其中内插區塊 包含至少一欲内插目標像素點；測試内插區塊之複數個相 〇 鄰區塊所對應之複數個向量之可信度；接著，選擇至少一 可信度咼之較佳向量，依據較佳向量產生至複數個權重 值；以及接收複數個向量所對應的像素值，根據複數個權 重值與複數個像素値，決定欲内插目標像素點之像素值。 201101841 【實施方式】 以下之說明將舉出本發明之數個較佳的示範實施 例，例如：各種電子電路、元件以及相關方法。熟悉本領 域者應可理解，本發明可採用各種可能的方式實施，並不 限於下列示範之實施例或實施例中的特徵。另外，眾所知 悉之細節不再重覆顯示或贅述，以避免模糊本發明之重 點。 ◎ 再者，本發明影像產生裝置、自適應移動補償内插法 (Adaptive blending MCI)與其裝置的輸入影像可為圖框 (frame)或圖場（fieId)。而為簡化說明，以下實施例均以圖 框為例作說明。 第1圖顯示本發明影像產生裝置之一實施例的架構方 塊圖。參考第1圖，本發明影像產生裝置1〇包含有一向 量估算裴置（Motion Vector estimation circuit)l 1 及一移動 補償内插裝置（Motion compensation interp〇lati〇n circuit)12。 〇 該向量估算裝置21〇之運作方式係將原始輸入圖框區 分成數個區塊（Macro block),每一個區塊均由Ν * Μ個 點所組成（比如說8 * 8、或8 * 16)，再以這些區塊來作移 動估計（motion estimation，ME)，在欲比對圖框中（例如前 一張圖框）找尋相似度最高的位置，其相對位置即為該區塊 所對應的移動向量。然後，尋找每一區塊所對應的向量， 就可以得到相鄰兩張畫面之間的所有向量群。 一實施例，向量估算裝置11接收一目前圖框Fi及— 201101841 先月J圖框FG將該目前圖框F1及先前圖框FG分為同樣數 =的區4再比較_目標區塊與先前圖框之—預設搜 尋品域t算出多個移動估計誤差值，並比較該些移動估 計誤差值，以得至丨丨马· n描广& μ目標區塊之移動向量MV及與节目才》 區塊相對應的多個移動估計誤差值。之後，不斷重複上^ 動作將整張圖框處理完後，向量估算裝置 相鄰兩張畫面之間的所有向量MV。 ❹ Ο 須注意，本發明並不限於此，另一實施例中，亦可比 2目心區塊Bi與目前圖框F1之一預設搜尋區域，以計201101841 VI. Description of the Invention: [Technical Field] The present invention relates to a Frame Rate Up-Conversion (FRC) technique, and more particularly to an Adaptive Motion MPI and its apparatus. [Prior Art] Frame rate up-conversion (FRC) o ^ is used to increase the picture frequency of the video source. There are various applications in the industry. For example, it is applied to low-order elements. Network video transmission to save bandwidth, applied to convert picture source with 25Hz picture frequency to more complex frequency to reduce juddering phenomenon, and applied to hold-type liquid crystal display (liquid crystal display 'LCD) to avoid blurring of the face and get clearer image quality. Most of the picture frequency improvement techniques first use motion estimation (ME) to calculate the motion vector of the object, and then perform motion compensation interpolation (MCI) to interpolate the different frames (frame). ) moving objects. Most of the motion estimation methods are performed in a block manner, so one of the main purposes of the motion compensation interpolation technique is to reduce block artifacts caused by block interpolation. The surrounding block is mixed to reduce block noise, but the general method does not distinguish whether the surrounding block is broken. Some methods use the inverse of the sum of absolute error (sum of absolute 201101841 difference, SAD) to the surrounding block. The choice but the calculation formula is too complicated, and can not correctly reflect the correctness of the surrounding blocks. SUMMARY OF THE INVENTION An embodiment of the present invention provides a motion compensation interpolation apparatus including a weight value unit and a mixing unit. The weight value unit receives at least one frame, the frame includes a plurality of blocks, and an interpolation block is planned in the preset number of the blocks, wherein the interpolation block includes at least Inserting a target pixel, and receiving a plurality of vectors corresponding to the plurality of blocks adjacent to the interpolated block, testing the vectors, and calculating a plurality of weight values according to at least one vector tested. The mixing unit receives a plurality of pixel values corresponding to the plurality of vectors, and generates pixel values of the target pixel to be interpolated according to the plurality of weight values and the plurality of pixels. An embodiment of the present invention provides a motion compensation interpolation method, which includes the following steps: First, an interpolation block is planned in a frame, wherein the interpolation block includes at least one target pixel to be interpolated; and the test interpolation area is tested. The reliability of the plurality of vectors corresponding to the plurality of neighboring blocks of the block; then, selecting a preferred vector of at least one confidence level, generating a plurality of weight values according to the preferred vector; and receiving the plurality of vectors The corresponding pixel value determines the pixel value of the target pixel to be interpolated according to the plurality of weight values and the plurality of pixels. 201101841 [Embodiment] The following description will set forth several preferred exemplary embodiments of the present invention, such as various electronic circuits, components, and related methods. It will be appreciated by those skilled in the art that the present invention may be embodied in a variety of possible ways and is not limited to the features of the following exemplary embodiments or embodiments. In addition, well-known details are not repeatedly shown or described in detail to avoid obscuring the invention. ◎ Furthermore, the image generating device of the present invention, the adaptive blending MCI and the input image of the device thereof may be a frame or a fieId. For simplicity of description, the following embodiments are illustrated by taking a frame as an example. Fig. 1 is a block diagram showing the construction of an embodiment of an image generating apparatus of the present invention. Referring to Fig. 1, the image generating apparatus 1 of the present invention includes a Motion Vector estimation circuit 1 and a Motion Compensation Interp〇 circuit (12). The operation mode of the vector estimating device 21 is to divide the original input frame into a plurality of blocks (Macro blocks), each of which is composed of Ν * Μ points (for example, 8 * 8 or 8 * 16) ), and then use these blocks for motion estimation (ME), and find the position with the highest similarity in the frame to be compared (for example, the previous frame), and the relative position is corresponding to the block. Mobile vector. Then, looking for the vector corresponding to each block, you can get all the vector groups between two adjacent pictures. In one embodiment, the vector estimating device 11 receives a current frame Fi and - 201101841. The first frame F frame and the previous frame FG are divided into the same number = area 4 and then compared _ target block and previous figure The preset search domain t calculates a plurality of motion estimation error values, and compares the motion estimation error values to obtain the motion vector MV of the target block of the Hummer and the μ target block and the program. 》 Multiple motion estimation error values corresponding to the block. After that, after repeating the ^ action to process the entire frame, the vector estimation device compares all vector MVs between two adjacent frames. ❹ 须 It should be noted that the present invention is not limited thereto. In another embodiment, the search area may be preset to be compared with one of the two-eye block Bi and the current frame F1.
异出多個移動估古+ -½ Μ a -P* I β -、差值及求出目標區塊之移動向量 MV。 移動補償内插裝置12參考該些移動向量卿進行補 ，處理’且用以將圖框進行内插，以產生至少—輸出圖 1。本發明一實施例之移動補償内插裝置12在運作時， =圖框中以預設大小之制區塊DB來進行内插處理。 例而δ ’如第1B、1C、1D圖所示，内插裝置12於ib 圖中’ 字型的四個區塊作為偵測區塊db來進行處 理’接著，處理完畢後再向右移動於第lc圖所示之位置 進行處理。當然，本發明並不限於僅採用此田字型四區塊 :進行處理’亦可採用其他數目之區塊大小來處理。再 ，偵測區塊DB之移動方向順序亦可依需求任意調整。 如第1D圖所示’偵測區塊DB中，田字型的四個區 塊為前述的區塊（Macro block)，並且每—區塊分別對應一 向量，如圖示中四個區塊分別對應MV1、MV2、mv3、 MV4等四個向量。而中間的灰色區塊是我們要處理的内插 7 201101841 區塊IB，該内插區塊IB同樣由N*M個點所組成。該内 插區塊IB和左上區塊相差半個區塊的距離，對於内插區 塊IB内的每一個欲内插的目標像素點（X，y)，我們可以找 到相鄰四個向量MV1、MV2、MV3、MV4，並根據四個向 量找到在前一或後一圖框所對應的像素點p 1、P2、p3、p4。 假設該目標像素點（X，y)與内插區塊IB左上頂點相距 (fx，fy)，則輸出像素Pout為 〇 PX2=^-fa)^Px+fi〇iP2 ^4=(1-»X^+>X^4Different motion estimates + -1⁄2 Μ a -P* I β -, difference, and the motion vector MV of the target block. The motion compensated interpolation device 12 performs the complement processing with reference to the motion vectors, and is used to interpolate the frames to produce at least the output map 1. In the operation of the motion compensation interpolation device 12 according to an embodiment of the present invention, the interpolation frame processing is performed by the block DB of the preset size in the frame. For example, as shown in FIGS. 1B, 1C, and 1D, the interpolation device 12 processes the four blocks of the 'font type in the ib diagram as the detection block db'. Then, after the processing is completed, it is moved to the right. Processed at the location shown in Figure lc. Of course, the invention is not limited to the use of only this type of block four blocks: processing ' can also be handled with other numbers of block sizes. Then, the order of the moving direction of the detection block DB can be arbitrarily adjusted according to requirements. As shown in FIG. 1D, in the detection block DB, the four blocks of the field type are the aforementioned blocks (Macro block), and each block corresponds to a vector, such as four blocks in the figure. Corresponding to four vectors such as MV1, MV2, mv3, and MV4. The middle gray block is the interpolation IB 201101841 block IB, which is also composed of N*M points. The distance between the interpolated block IB and the upper left block is half the block. For each target pixel (X, y) to be interpolated in the interpolated block IB, we can find the adjacent four vectors MV1. , MV2, MV3, MV4, and find the pixel points p 1 , P2 , p3 , p4 corresponding to the previous or next frame according to the four vectors. Assuming that the target pixel point (X, y) is spaced from the upper left vertex of the interpolated block IB (fx, fy), the output pixel Pout is 〇PX2=^-fa)^Px+fi〇iP2^4=(1-» X^+>X^4
= (1 - ^) X + ^ X ϋ)(ΐ-刷 +(1-办)_ +_ 一规·_ = WxP,+W2P2+W,Pi+W,P, QVx+W2+W,+Wa=\) (1.1) 其中，Pn為在前一或後一圖框中，距該點位置向量MVn所 對應的像素值，且n=l〜4。 A為在灰色内插區塊IB中，橫向距離除以内插區塊 Q IB長度的值。 為在灰色内插區塊iB中，縱向距離除以内插區塊 IB寬度的值。 舉例而言，假設相鄰的四個向量為（〇,〇)、（〇,〇)、（6,〇)、 (6,0)，並且欲内插之目標像素點（x,y)所對應的像素值為= (1 - ^) X + ^ X ϋ)(ΐ-刷+(1-办)_ +_一规·_ = WxP, +W2P2+W, Pi+W, P, QVx+W2+W,+ Wa=\) (1.1) where Pn is the pixel value corresponding to the position vector MVn in the previous or next frame, and n=l~4. A is the value of the lateral distance divided by the length of the interpolated block Q IB in the gray interpolated block IB. In the gray interpolated block iB, the longitudinal distance is divided by the value of the interpolated block IB width. For example, suppose the four adjacent vectors are (〇, 〇), (〇, 〇), (6, 〇), (6, 0), and the target pixel (x, y) to be interpolated Corresponding pixel value
Pl = 100、P2=200、 P3=300 與 P4=4〇〇，且假設該内插目 標像素點（X，y)的fx = 4、fy = 8，而内插區塊IB長寬為 16*16 »則輸出像素值p〇ut為 201101841Pl = 100, P2=200, P3=300, and P4=4〇〇, and assume that the interpolated target pixel (X, y) has fx = 4, fy = 8, and the interpolated block IB has a length and width of 16 *16 »The output pixel value p〇ut is 201101841
Pout = (1-8/16)*(1-4/16)*100 + (l-8/16)*(4/16)*200 + (8/16)*(l-4/16)*300 + (8/16)*(4/l 6)*400 = 225 依此方式，重覆運算可求出内插區塊IB中每一像素 的輸出像素值Pout，進而求出整張内插圖框之全部像素 值。然而，並非每一個鄰近區塊之向量皆具有高的可信 度’因此需要根據向量之可信度高低來分配各向量之權重 大小。 第2A圖顯示本發明一實施例之自適應移動補償内插 裝置12。該内插裝置12包含有一權重值單元（Adaptive Weighting Unit)121 與一混成單元（Blending Unit)122。權 重值單元121接收一内插區塊IB之相鄰至少一向量MV， 其中該内插區塊IB包含一内插目標像素點（x，y)，且經處 理後產生至少一權重值W。混成單元122接收相鄰至少一 向量MV所對應的至少一像素值P，根據該至少一權重值 W處理該至少一像素値P，以產生該内插目標像素點之輸 出像素值Pout。 如第2A圖之示例，權重值單元12丨接收一内插區塊 IB之相鄰四向量mvi、MV2、MV3、MV4，其中内插目標 像素點（x，y)之内插位置為（fx，fy)，於相鄰四向量中求出偏 好的向量MV且經處理後產生權重值wi、W2、W3'W4 至混成單元122處理。 混成單元122接收相鄰四向量所對應的像素值ρι、 P2、P3、P4，且根據該權重值W1、W2、W3、W4處理該 像素値PI、P2、P3、P4,以產生該内插點像素之輸出像素 值OP 〇 9 201101841 須注意’權ί值單元121可輪出適當的權重值w來偏 向可信度較高的向量MV。舉例而言，如果向量_的 可信度比其他的向量要高，則增強該向量對應像素值Η 的權重值W1。 >舉例而言，可以利用下列较置與方法來決定向量的可 k度，並求出適當的轉換值。一實施例，如第2B圖所示， 權重值單元121包含有細節測試單元121a、移動估計誤差 0 值測試單元l21b、向量一致性測試單元121c、以及一找查 表（Look up table)12lde如果該向量mv所對應的區塊係 具有較高的細節紋理（Highdetai丨）、較低的移動估計誤差值 (Low SAD)、以及較高的向量一致性（mgh Mv(M〇^ngPout = (1-8/16)*(1-4/16)*100 + (l-8/16)*(4/16)*200 + (8/16)*(l-4/16)* 300 + (8/16)*(4/l 6)*400 = 225 In this way, the repeated calculation can find the output pixel value Pout of each pixel in the interpolation block IB, and then find the whole inner illustration. The full pixel value of the box. However, not every vector of neighboring blocks has a high degree of confidence. Therefore, it is necessary to allocate the weight of each vector according to the reliability of the vector. Fig. 2A shows an adaptive motion compensation interpolating device 12 in accordance with an embodiment of the present invention. The interpolation device 12 includes an Adaptive Weighting Unit 121 and a Blending Unit 122. The weight value unit 121 receives the adjacent at least one vector MV of an interpolated block IB, wherein the interpolated block IB includes an interpolated target pixel (x, y) and is processed to generate at least one weight value W. The mixing unit 122 receives at least one pixel value P corresponding to the adjacent at least one vector MV, and processes the at least one pixel 値P according to the at least one weight value W to generate an output pixel value Pout of the interpolation target pixel. As in the example of FIG. 2A, the weight value unit 12 receives an adjacent four vector mvi, MV2, MV3, MV4 of an interpolation block IB, wherein the interpolation position of the interpolation target pixel (x, y) is (fx) , fy), find the preferred vector MV in the adjacent four vectors and process the generated weight values wi, W2, W3 'W4 to the blending unit 122 for processing. The mixing unit 122 receives the pixel values ρι, P2, P3, and P4 corresponding to the adjacent four vectors, and processes the pixels 値PI, P2, P3, and P4 according to the weight values W1, W2, W3, and W4 to generate the interpolation. Point pixel output pixel value OP 〇9 201101841 It should be noted that the 'weight value unit 121 can rotate the appropriate weight value w to favor the vector MV with higher reliability. For example, if the reliability of the vector_ is higher than the other vectors, the weight value W1 of the corresponding pixel value Η of the vector is enhanced. > For example, the following comparison method can be used to determine the k-degree of the vector and find an appropriate conversion value. In an embodiment, as shown in FIG. 2B, the weight value unit 121 includes a detail testing unit 121a, a motion estimation error 0 value testing unit 112b, a vector consistency testing unit 121c, and a lookup table 12lde. The block corresponding to the vector mv has a high detail texture (Highdetai丨), a low motion estimation error value (Low SAD), and a high vector consistency (mgh Mv(M〇^ng)
Vector) coherence)其令之一或其組合時，則該向量為一可 度高的向量。 一實施例，可以利用調整内插位置fx、fy的值來調 整權重谊W，以使輸出像素值〇p偏向該向量所對 素值。 ：) 須注意，本發明決定向量可信度之裝置與方法並不限 於上述，可採用其他目前現有或未來可能發展出之裝置與 方法來判斷。 權重值單元121之測試運作方式如下： 1·細節測試單元12la進行細節測試（High如…夏 checkmg)。細節職單元121a測試每該向量區塊的細節 、文理其中，具有較南細節紋理的向量區塊，在找尋移動 估計誤差值時較能反映出正確的向量，因此須選擇具有較 10 201101841 來曰、丨# 塊。細節測試之運作可利用以下公式 來莖測，量測該向量區塊與該偏移右下一格 塊的差異絕對值，量測出之數值 2 大。公式如下： 敎、細卽紋理之値越 石(丨厂(尤，少）+1，少+1)丨) > 決⑽加奶 (1.2) b1〇ck為該區塊；F為圖框；thresh〇ldi為第一臨界值。 Ο ❹ 2.移動估計誤差值測試單元Ulb進行移動估計誤差 值測試（L〇W SAD decking)。移動估計誤差值測試單元 121b測量内插區塊所對應鄰近向量（如第1〇圖中 的移動估計誤差值，較小的誤差值表示該向量 的可信度越高。移動估計誤差值測試可利用以下公式來旦 測： 里 ㈣-释邊，”聊㈣祕咖 (13) 其中，block為該内插區塊，F〇為先前圖框；F1為 目前圖框，MVx、MVy為向量；threshold2為第二臨界值 3·向量一致性測試單元121c進行向量一致性測試 (MV(Moving Vector) coherence checking)。將任一向量八 別與其鄰近的向量進行測試，例如將向量MV 1分別與向 量MV2〜MV4進行測試，若待測向量MV1與周圍鄰近向 量相似’則該向量的可信度較高，則選擇偏向該向量。可 利用以下公式來量測向量一致性： 11 201101841 (1.4) Z \neighborMV -MV\ < threshold^ neighborhood 其中，neighborhood指鄰近之區塊，MV指待測向量； «ekWorMV指鄰近待測向量的向量；threshold3為第三臨 界值。 須注意，當一向量所對應的區塊具有較高的細節紋 理、較低的移動估計誤差值、以及較高的向量一致性時， 則該向量為一可信度高的向量。當然，本發明對於向量偏 oVector) Coherence) When one of its orders or a combination thereof, the vector is a highly vector. In one embodiment, the weights W can be adjusted by adjusting the values of the interpolation positions fx, fy such that the output pixel value 〇p is biased toward the pixel value of the vector. :) It should be noted that the apparatus and method for determining the vector confidence of the present invention are not limited to the above, and may be judged by other devices and methods that are currently available or may be developed in the future. The test operation mode of the weight value unit 121 is as follows: 1. The detail test unit 12la performs detailed test (High such as ... summer checkmg). The detail unit 121a tests the details and the texture of each of the vector blocks. The vector block with the south detail texture can reflect the correct vector when looking for the motion estimation error value, so it must be selected to have a comparison with 10 201101841. , 丨# block. The operation of the detail test can use the following formula to measure the absolute value of the difference between the vector block and the right next block of the offset, and the measured value is 2 large. The formula is as follows: 敎, fine 卽 texture of 値石(丨厂(尤,少)+1,少+1)丨) > 决(10)加奶(1.2) b1〇ck is the block; F is the frame ;thresh〇ldi is the first critical value. Ο ❹ 2. The motion estimation error value test unit Ulb performs a motion estimation error value test (L〇W SAD decking). The motion estimation error value test unit 121b measures the neighboring vector corresponding to the interpolation block (such as the motion estimation error value in FIG. 1 , and the smaller error value indicates that the reliability of the vector is higher. The motion estimation error value test may be Use the following formula to test: 里(四)- 释边,“聊(四)秘咖(13) where block is the interpolated block, F〇 is the previous frame; F1 is the current frame, MVx and MVy are vectors; Threshold 2 is a second threshold value 3 · vector consistency test unit 121c performs a vector consistency test (MV (Moving Vector) coherence checking). Test any vector eight and its neighboring vectors, for example, vector MV 1 and vector respectively MV2~MV4 are tested. If the vector MV1 to be tested is similar to the surrounding neighboring vector', then the reliability of the vector is higher, then the vector is biased. The following formula can be used to measure the vector consistency: 11 201101841 (1.4) Z \neighborMV -MV\ < threshold^ neighborhood where, refers to the neighboring block, MV refers to the vector to be tested; «ekWorMV refers to the vector adjacent to the vector to be tested; threshold3 is the third critical value. When the block corresponding to a vector has a higher detail texture, a lower motion estimation error value, and a higher vector consistency, the vector is a vector with high reliability. Of course, the present invention Vector bias o
向的判斷’可依據設計者之需求任意調整。例如可僅採用 上述任一、或任二測試方式來進行判斷、或是另外增加測 試的方式、或是採用其他現有或未來發展出之測試方式進 行判斷。 接著，對於可信度較高的向量，可以利用找查表121d 調整内插位置fx、fy之數值成為fx，、fy，，再調整權重 値W1 ^2、W3、W4成為新的權重值wi’、W2，、W3，、 W4，達成偏向其所對應的像素值之效果，如第3圖所示。 方程式如下： ~^^+ψ2'Ρ2+Ψ3'ρ3+ίν4'ρ (K+w2'+fv3'+fr4'=^ (1.5) 舉例而言.如里a θ 1 ▲ 。·如果向量MV4通過「細節測試」、「移動 一：差值測s式」、與「向量一致性測試」，則該向量具有 數信 我們可以透過找查表121d轉換内插位置fx、fy 的方式來得到較大的向量W4，的值。較大的向量w4, 在混成公式中可 J 乂反應較多P4的值，而P4是向量MV4 12 201101841 所對應的像素值’因此可透過轉換内插位置fx、fy數値 的方式來偏向可信度高的向量MV4。 内插位置fx，fy數値的轉換可利用找查表121d内建 之圖表（如第4A、4B圖所示）來轉換，說明如下： 若為原線性内插（不偏向任一向量）：於第4A 4b圖 中’選擇情況X1 +情況γ 1兩轉換表。 若需偏向MV1 (左上向量）：第4A、4B圖中，選擇情 況X2 +情況Y2兩轉換表。 〇 若需偏向MV2 (右上向量）：第4A、4B圖中，選擇情 況X3 +情況Y2兩轉換表。 若需偏向MV3 (左下向量）：第4A、4B圖中，選擇情 況X2 +情況Y3兩轉換表。 若需偏向MV4(右下向量）：第4A、4B圖中，選擇情 況X3 +情況Y3兩轉換表。 舉例而言，假設相鄰的四個向量為（〇,〇)、（〇,〇)、（6,〇)、 (6,0)，並且欲内插之目標像素點（x，y)所對應的像素值為 〇 P1 = 1〇〇'P2=200、P3=300 與 P4=400,且假設該像素點（x,y) 的内插位置數値fx = 4、fy = 8，而内插區塊長寬為16*16。 假設經過權重值單元121測試後可信之向量為向量MV4， 因此選擇情況X3 +情況Y3兩轉換表轉換使fx、fy偏 向向量MV4,得到&，= 8、&，= 14。將此數値帶入公式（1.5)， 可得到偏向向量MV4輸出像素值p〇ut為The judgment of the direction can be arbitrarily adjusted according to the needs of the designer. For example, the determination may be made by using only one of the above, or any of the two test methods, or by adding another test, or by using other existing or future developed test methods. Then, for the vector with higher reliability, the value of the interpolation positions fx and fy can be adjusted to become fx, fy by using the lookup table 121d, and the weights W1^2, W3, and W4 can be adjusted to become the new weight value wi. ', W2, W3, and W4, achieve the effect of biasing the pixel value corresponding to it, as shown in Figure 3. The equation is as follows: ~^^+ψ2'Ρ2+Ψ3'ρ3+ίν4'ρ (K+w2'+fv3'+fr4'=^ (1.5) For example, if a θ 1 ▲ · If the vector MV4 passes "Detail test", "Move one: difference test s", and "vector conformance test", the vector has a number of letters. We can find the larger position by converting the interpolation positions fx and fy through the lookup table 121d. The value of vector W4, the larger vector w4, in the blending formula, J 乂 reflects more P4 values, and P4 is the pixel value corresponding to vector MV4 12 201101841 'so can be converted by interpolation interpolation position fx, fy The number of ways to bias the vector MV4 with high reliability. The conversion of the interpolation position fx, fy number 可 can be converted by using the built-in chart of the lookup table 121d (as shown in Figures 4A and 4B), as follows: If the original linear interpolation (not biased to any vector): in the 4A 4b diagram, 'select the situation X1 + case γ 1 two conversion table. If you need to bias MV1 (upper left vector): 4A, 4B, select the situation X2 + case Y2 two conversion table. 〇 If you need to bias MV2 (upper right vector): In the 4A, 4B picture, select case X3 + case Y2 two conversion table. If you need to bias MV3 (lower left vector): 4A, 4B, select case X2 + case Y3 two conversion table. If you need to bias MV4 (lower right vector): 4A, 4B, select case X3 + case Y3 two For example, assume that the four adjacent vectors are (〇, 〇), (〇, 〇), (6, 〇), (6, 0), and the target pixel to be interpolated (x, y) The corresponding pixel values are 〇P1 = 1〇〇'P2=200, P3=300 and P4=400, and the number of interpolation positions of the pixel (x, y) is assumed to be 値fx = 4, fy = 8 The interpolated block length and width are 16*16. It is assumed that the trusted vector after the weight value unit 121 is tested is the vector MV4, so the selection case X3 + case Y3 two conversion table conversions make fx, fy biased to the vector MV4, and get & , = 8, &, = 14. Bring this number 公式 into the formula (1.5), you can get the bias vector MV4 output pixel value p〇ut
Pout=(l-14/16)*(l-8/l6)*i00+ (1-14/16)*(8/16)*200+(ΐ4/ΐ6)*(ΐ_8/ΐ6)*300+(14/16)*(8/16 13 201101841 )*400=0·0625*100+0.0625*200+0.4375*300+0.4375*400= 325 可得到權重俊 W1，=〇.0625、W2，=0.0625、 W3’=0.4375、W4’ = 0.4375，其中 Wl，+ W2，+ W3’ + W4’ = l。 依此方式，重複以偏向向量MV4之方式計算出整個 内插區塊BI中的全部目標像素點（x，y)之輸出像素值 Pout ’即可得到整張的内插圖框，而產生第1A圖影像產 生裝置10之輸出圖框OF。由於此輸出圖框〇F係依據可 〇 信度較高之向量内插出，因此其可消除習知技術之塊狀雜 訊（block artifacts)，且較先前技術利用周邊區塊混成的方 式減低塊狀雜訊之方式較為簡單、且本發明之可信度較 高’可達成正確反應週邊區塊正確性之功效。 須注意，本發明影像產生裝置與方法中之一實施例 中，假如鄰近四個向量均無法通過可信度測試，則仍選擇 原線性内插的方式來混成，如果有兩個以上的向量均為可 信度高的向量，則可以選擇其中之一來處理。 Q 須注意，上述說明均係採用四個向量來處理，當然本 發明並不限於此，亦可採用其他數目的向量來處理，向量 取用之數目可等於N，其中N為正整數小於無限大。 再者，第5圖顯示本發明一實施例之一種移動補償内 插方法之流程圖。該方法包含有下列步驟： 步驟S502 :開始。 步驟S504 :於—圖框規劃一内插區塊，其中該内插區 塊包含至少一欲内插目標像素點。 步驟S506:測試該内插區塊之複數個相鄰區塊所對應 14 201101841 之複數個向量之可信度。 步驟S508:選擇至少一可信度高之較佳向量，依據該 較佳向量產生複數個權重值。 / 步驟S5 10:接收該複數個向量所對應的像素值，根據 該複數個權重值與該複數個像素値，決定該欲内插目標像 素點之像素值。 步驟S512:結束。 0 以上雖以實施例說明本發明’但並不因此限定本發明 之範圍，只要不脫離本發明之要旨，該行業者可進行各種 變形或變更。 【圖式簡單說明】 第1A圖顯示本發明一實施例之影像產生裝置之示意 圖。 第1B圖顯示本發明—實施例之圖框區塊内插處理之 示意圖。 Ο 第1 C圖顯示本發明另一實施例之圖框區塊内插處理 之示意圖。 第1D圖顯示本發明另一實施例之圖框區塊内插處理 之示意圖。 第2A圖顯示本發明一實施例之自適應移動補償内插 裝置之示意圖。 第2B圖顯示本發明一實施例之權重值單元之示意圖。 _第3圖顯示本發明另—實施例之圖框區塊内插處理之 示意圖。 15 201101841 第4A圖顯示本發明一實施例之向量權重判斷處理之 示意圖。 第4B圖顯示本發明另一實施例之向量權重判斷處理 之示意圖。 第5圖顯示本發明一實施例之自適應移動補償内插方 法之流程圖。 【主要元件符號說明】 〇 10 影像產生裝置 11 向量估算電路 12 移動補償内插電路 121 權重値單元 122 混成單元 121a 〜121c 測試單元 121d 找查表 16Pout=(l-14/16)*(l-8/l6)*i00+ (1-14/16)*(8/16)*200+(ΐ4/ΐ6)*(ΐ_8/ΐ6)*300+( 14/16)*(8/16 13 201101841 )*400=0·0625*100+0.0625*200+0.4375*300+0.4375*400= 325 You can get weights W1,=〇.0625, W2,=0.0625, W3'=0.4375, W4' = 0.4375, where Wl, + W2, + W3' + W4' = l. In this way, the output pixel value Pout ' of all the target pixel points (x, y) in the entire interpolated block BI is calculated by repeating the deviation vector MV4 to obtain the entire inner frame, and the first frame A is generated. The output frame OF of the image generating device 10 is shown. Since the output frame 〇F is interpolated according to a vector with higher reliability, it can eliminate block artifacts of the prior art and is reduced by the prior art by using the surrounding block. The method of block noise is relatively simple, and the credibility of the present invention is high', and the effect of correctly correcting the correctness of the surrounding block can be achieved. It should be noted that in one embodiment of the image generating apparatus and method of the present invention, if the neighboring four vectors cannot pass the credibility test, the original linear interpolation method is still selected to be mixed, if there are more than two vectors. For highly reliable vectors, you can choose one of them to handle. Q It should be noted that the above description is performed by using four vectors. Of course, the present invention is not limited thereto, and other numbers of vectors may be used for processing. The number of vector accesses may be equal to N, where N is a positive integer smaller than infinity. . Furthermore, Fig. 5 is a flow chart showing a method of motion compensation interpolation according to an embodiment of the present invention. The method comprises the following steps: Step S502: Start. Step S504: An interpolation block is planned in the frame, wherein the interpolation block includes at least one pixel to be interpolated. Step S506: Test the reliability of the plurality of vectors corresponding to the multiple adjacent blocks of the interpolated block. Step S508: Select at least one preferred vector with a high degree of confidence, and generate a plurality of weight values according to the preferred vector. / Step S5 10: receiving the pixel value corresponding to the plurality of vectors, and determining the pixel value of the pixel to be interpolated according to the plurality of weight values and the plurality of pixels. Step S512: End. The present invention is described by way of example only, and the scope of the invention is not limited thereto, and various modifications and changes can be made by those skilled in the art without departing from the scope of the invention. BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1A is a schematic view showing an image generating apparatus according to an embodiment of the present invention. Fig. 1B is a view showing the interpolation processing of the block block of the present invention. Ο Figure 1C is a diagram showing the interpolation processing of the block block of another embodiment of the present invention. Fig. 1D is a view showing the interpolation processing of the block block of another embodiment of the present invention. Fig. 2A is a diagram showing an adaptive motion compensation interpolating apparatus according to an embodiment of the present invention. Figure 2B is a diagram showing the weight value unit of an embodiment of the present invention. Fig. 3 is a view showing the interpolation processing of the frame block of another embodiment of the present invention. 15 201101841 Fig. 4A is a diagram showing the vector weight determination processing of an embodiment of the present invention. Fig. 4B is a view showing the vector weight judging process of another embodiment of the present invention. Fig. 5 is a flow chart showing an adaptive motion compensation interpolation method according to an embodiment of the present invention. [Description of main component symbols] 〇 10 Image generating device 11 Vector estimating circuit 12 Motion compensation interpolation circuit 121 Weight 値 unit 122 Mixing unit 121a to 121c Test unit 121d Lookup table 16